Hair Cortisol Concentration, Weight Loss Maintenance and Body Weight Variability: A Prospective Study Based on Data From the European NoHoW Trial

Several cross-sectional studies have shown hair cortisol concentration to be associated with adiposity, but the relationship between hair cortisol concentration and longitudinal changes in measures of adiposity are largely unknown. We included 786 adults from the NoHoW trial, who had achieved a successful weight loss of ≥5% and had a body mass index of ≥25 kg/m2 prior to losing weight. Hair cortisol concentration (pg/mg hair) was measured at baseline and after 12 months. Body weight and body fat percentage were measured at baseline, 6-month, 12-month and 18-month visits. Participants weighed themselves at home ≥2 weekly using a Wi-Fi scale for the 18-month study duration, from which body weight variability was estimated using linear and non-linear approaches. Regression models were conducted to examine log hair cortisol concentration and change in log hair cortisol concentration as predictors of changes in body weight, change in body fat percentage and body weight variability. After adjustment for lifestyle and demographic factors, no associations between baseline log hair cortisol concentration and outcome measures were observed. Similar results were seen when analysing the association between 12-month concurrent development in log hair cortisol concentration and outcomes. However, an initial 12-month increase in log hair cortisol concentration was associated with a higher subsequent body weight variability between month 12 and 18, based on deviations from a nonlinear trend (β: 0.02% per unit increase in log hair cortisol concentration [95% CI: 0.00, 0.04]; P=0.016). Our data suggest that an association between hair cortisol concentration and subsequent change in body weight or body fat percentage is absent or marginal, but that an increase in hair cortisol concentration during a 12-month weight loss maintenance effort may predict a slightly higher subsequent 6-months body weight variability. Clinical Trial Registration ISRCTN registry, identifier ISRCTN88405328.


INTRODUCTION
Worldwide, obesity is one of the leading preventable causes of both morbidity and mortality (1)(2)(3). While existing approaches are generally effective for initial weight loss (WL), behavioural interventions have shown quite limited effects on long-term weight loss maintenance (WLM) (4). Thus, understanding the potential causes or predictors of WLM and weight regain can have important therapeutic implications. Dietary relapse and/or a reduction in physical activity are the obvious courses of weight regain, but there are likely numerous genetic, behavioural, metabolic and hormonal factors involved (5)(6)(7)(8).
While the underlying mechanisms are not fully understood, a link between perceived stress and obesity has been suggested for decades (9). It has also been reported that perceived stress predicts weight regain after clinically significant WL (10). Several factors and pathways could potentially explain this relationship, including that psychological stress can affect the hypothalamic-pituitary-adrenal (HPA) axis and lead to an increase in cortisol secretion (11). This increase in cortisol may promote weight regain through an increase in appetite (8), but it has also been suggested that cortisol can directly promote fat disposition, as seen in Cushing's disease where frequent symptoms are weight gain, increased body fat (BF) and central obesity (12).
In addition to an association between perceived stress and weight regain, some studies have shown tendency towards irregular eating patterns among individuals reporting higher levels of stress, an eating behaviour referred to as 'stress eating' (13)(14)(15). This could theoretically lead to a higher degree of fluctuation in caloric intake and thereby promote a higher body weight variability (BWV), which by itself has been suggested as a predictor of future weight gain (16)(17)(18). However, to the best of our knowledge, an association between stress and BWV has neither been explored using self-reported nor biological measures of stress.
Measurement of cortisol in blood, saliva or urine are often used as biological markers of stress. However, an important limitation to these measures is that they provide information on current cortisol level, and consequently reflects acute, rather than chronic, stress. In contrast to these measures, hair cortisol concentration (HCC) is a biomarker for assessing chronic stress, by reflecting cortisol exposure during the period of hair growth (19). While previous studies have shown inconsistent associations between HCC and perceived stress (20), the measure has been shown to align with saliva cortisol production measured continuously over a corresponding onemonth period (19).
Several studies have shown HCC to be associated with adiposity at a cross-sectional level (21)(22)(23)(24), but the relationship between HCC and long-term WLM, or longitudinal weight change in general, is largely unknown and as a consequence the direction of causality has not yet been determined.
The present study is an ancillary study based on data from the European NoHoW trial, where the aim was to test the efficacy of an evidence-based digital toolkit, targeting self-regulation, motivation, and emotion regulation, on WLM among British, Danish, and Portuguese adults (25). Using the NoHoW data, the aim of the present study was to examine the longitudinal associations between HCC, WLM and body weight variability. More specifically, we examined: 1) the association between baseline HCC and subsequent changes in body weight (BW) (primary outcome), change in BF%, and BWV (secondary outcomes) during 6, 12 and 18 months of follow-up, 2) the association between 12-month change in HCC and 12-month develop in outcome measures during the same period, and 3) the association between initial 12-month change HCC and subsequent development in outcome measures between month 12 and month 18.

Study Population
A detailed description of NoHoW can be found elsewhere (25). The trial was registered with the ISRCTN registry (ISRCTN88405328). NoHoW was a randomised controlled trial testing the efficacy of an information and communications technology (ICT)-based toolkit to support WLM in the United Kingdom (Leeds), Denmark (Copenhagen), and Portugal (Lisbon). At baseline examinations between March 2017 and March 2018, all participants were randomly allocated to one of four arms: (1) self-monitoring only (self-weighing and activity tracker), (2) self-regulation plus motivation, (3) emotion regulation, or (4) combined selfregulation, motivation, and emotion regulation. Participants were followed during a 12-month period for change in BW (primary trial outcome), body composition, biomarkers, dietary intake, physical activity, sleep, and psychological mediators/moderators of WLM (secondary trial outcomes). In addition to the primary follow-up visit at month 12, measurements were collected at month 6 and month 18. Participants were 18 years or older, had achieved a verified and clinically significant WL of ≥5% within the 12 months prior to inclusion, and had a BMI of ≥25 kg/m 2 before losing weight. The exclusion criteria were as follows: achieved WL due to illness or surgical procedures; pregnancy or breastfeeding; involvement in other research intervention studies that confound with the aims of the intervention; inability to follow written material or telephone conversations in the English, Danish, or Portuguese language (depending on the trial centre); diagnosis of an eating disorder; diagnosis of any condition that may interfere with increasing mild to moderate physical activities and that is unstable (i.e., untreated or unable to be controlled by medication); recent diagnosis of type 1 diabetes; extensive travel plans (e.g., more than four weeks); or living in the same household as existing participant in the trial (25). A total of 1,627 participants were enrolled in the NoHoW trial.
The present ancillary study was based on information collected at baseline, 6, 12 and 18-month follow-up visits. We further excluded participants with missing information on baseline HCC (n=496), no follow-up information on the primary outcome (BW) (n=157), and missing information on selected covariates (n=188). A total of 786 participants had information on HCC, selected covariates and information on BW from at least one follow-up visit (month 6 [n=773], month 12 [n=713] and month 18 [n=668]). A slightly lower sample size was used for the secondary outcomes (BF% and BWV).

Hair Cortisol Concentration
Hair samples were cut from the posterior vertex as close to the scalp as possible. Between 10-30 mg of the hair from 2 cm closest to the scalp was cut in small pieces and dissolved in 1 mL methanol and incubated at ultrasound sonication for 30 minutes, followed by 18 hours at 52°C in a shaking incubator (300 rpm.). Hair samples were not collected among participants with less than 2 cm of hair. The methanol was transferred to a new tube and evaporated to dryness under a stream of nitrogen at 45°C. Dried samples were stored at -20°C until analysis. Before analysis, samples were re-dissolved in 500µl PBS pH 8 and centrifuged at 2000 rpm for 2 min. Reconstituted samples were analysed with a cortisol ELISA assay (Alpco.com). The measurement range of the assay was 0.15-50 ng/mL. We used three levels of control specimens to show the inter-assay variation as the inter-assay precision might be different depending on the measurements level. The inter-assay variation was 16.7% (mean 0,22 ng/mL), 11,4% (mean 9,8 ng/ mL) and 7.7% (mean 49.2 ng/mL). HCC was expressed as pg cortisol/mg hair.
We aimed to analyse 10-30 mg of hair. However, especially among the male participants, the collected samples did not consistently contain enough hair to meet this criterion. Thus, as we found a direct association between weight of the hair sample and HCC, weight of hair samples (mg) was included in sensitivity analyses."

Body Weight and Body Fat Percentage
At baseline and follow-up visits, BW was measured to the nearest 0.1 kg and height was measured at baseline to the nearest 0.1 cm using the Seca 704s (SECA, Hamburg, Germany) combined stadiometer and electronic scale. Body composition was estimated from bioelectrical impedance using the ImpediMed SFB7 device (ImpediMed, Inc, Sydney, Australia) (software version 5.4.0), following the manufacturer's instructions. We used the Moissl BMI modification of the mixture theory equations to determine BF%, a method that has been found appropriate over a wide range of body compositions (26). Changes in BW and BF% between baseline and 6, 12 or 18 months and between month 12 and 18 (follow-up values minus baseline values) were included in analyses as continuous variables.

Body Weight Variability
In addition to the weight measurements conducted at the official centre visits, all participants were provided with a Fitbit Aria smart scale (Fitbit Inc, San Francisco, CA, USA). This scale was linked to a personalised Fitbit account and the data were retrieved via a Fitbit application programming interface (API) to an online data hub. The Fitbit Aria scale has been shown to have excellent agreement with a calibrated research grade SECA 769 scale (27). All participants were instructed to weigh themselves on the Fitbit Aria scale at least twice per week for the duration of the study. From this, BWV was estimated from the Root Mean Square Error (RMSE) (28)(29)(30). Moreover, we included an additional measure, which we have previously developed to overcome limitations of commonly used approaches (including assumption of linearity in weight trajectory), termed the non-linear mean deviation (NLMD) method. Detailed information on the methods used to calculate BWV has been published elsewhere (30,31). In brief, RMSE was estimated by taking the mean square of the relative residual error in the linear relationship between BW and time. The NLMD was calculated by fitting a locally estimated scatterplot smoothing (LOESS) regression to the BW data which acts as a smoother. The fit of the regression was then subtracted from the observed BW and the relative mean deviation from the non-linear trend was calculated. RMSE and NLMD from baseline to month 6, 12 and 18 and between months 12 and 18 were included in the analyses as continuous variables (%).

Covariates
Participants provided verified information on WL during the 12 months prior to inclusion at baseline (by a health professional, WL counsellor/friend, WL programme record booklet, diary, smartphone app, or before/after photographs), and WL was included in the analyses as a continuous variable (kg). At baseline, all participants were allocated to an intervention arm and this information was included as a categorical variable: (1) selfmonitoring only (self-weighing and activity tracker), (2) selfregulation plus motivation, (3) emotion regulation, or (4) combined self-regulation plus motivation and emotion regulation. Participants were asked to report their smoking status, and were categorised as current, previous or never smokers. Information was also collected on alcohol consumption, which was included using the following 6 categories: every day, 5-6 times a week, 3-4 times a week, twice a week, once a week, and less than once a week. Information on highest level of education was provided and categorised according to the International Standard Classification of Education (ISCED) (32) as high, medium, low, or other (including educations not classified by ISCED. Age (continuous variable) and sex was also included. Objectively measured physical activity was measured using the Fitbit Charge 2 (33,34). The device was updated to the latest firmware and worn on the nondominant wrist of the participant. Participants were instructed to wear the Fitbit Charge 2 for the duration of the study, apart from during water activities (e.g., showering) and when charging the device. In the present study, we used 14 days of sleep data recorded from day 3 to day 16 of the intervention. Days 1 and 2 were excluded to ensure all devices had been properly set up. When no heart rate data were available from the Fitbit, we considered it as non-wear time. To avoid loss of data due to connectivity issues, gaps of less than 10 minutes were imputed with the average of the last measured and the next observed heart rate. Minute-level data were aggregated to hourly data and missing time was determined per hour. Total number of steps were divided by the number of minutes the device was worn, on the assumption that data missing within each hour were most representative of missing data. Hours with more than 30 minutes of missing data were removed from the data. Next, hourly averages were summed per day, a minimum of 21 valid hours were required for a valid day. Total steps were then averaged across the 14-day period if at least 6 valid days and 2 weekend days were available. From this, physical activity measured as average steps/day (continuous variable) was included. Lastly, information on centre/country of residence was included.
In sensitivity analyses, information on sleep habits (sleep duration and sleep onset variability) and psychosocial stress were added. Objectively measured sleep was also collected using the Fitbit Charge 2. Sleep duration and sleep onset was assessed across the 14 days close to baseline examinations using a previously described approach (35,36). Mean daily sleep duration was calculated and included in analyses as a categorical variable: <6, 6-<7, 7-<8, 8-<9, and ≥9 hours. Moreover, mean sleep onset was assessed by identifying the beginning of each main sleep period in minutes from midnight (e.g., 23:00 = −60 minutes and 01:00 = 60 minutes). From this, variability in sleep onset was estimated for each individual using the standard deviation across all nights recorded and included as a continuous variable (hours). Psychosocial stress was assessed with the short version of the perceived stress scale (PSS) (37,38). This scale has the following four items that focus on the assessment of stress and coping over the preceding month: 'How often have you felt that you were unable to control the important things in your life?' 'How often have you felt confident about your ability to handle your personal problems?' 'How often have you felt that things were going your way?' and 'How often have you felt difficulties were piling up so high that you could not overcome them?'. Responses were made on a 5-point Likert scale (never, rarely, sometimes, often, and very often). The items were then summed to give a total perceived stress score with a range of 0 (least stressed) to 16 (most stressed), which was included as a continuous variable.

Statistical Analyses
All statistical analyses were guided by an analysis plan developed prior to analysing the data (Supplementary Text 1). The sample size of the NoHoW trial was determined with the main purpose of having sufficient statistical power to detect a potential effect of the intervention on change in BW (25). As the present results represent an ancillary study, the sample size was not determined specifically for the purpose of these analyses. However, the preestablished sample size of 786 individuals available for the current study gave approximately 80% power to detect correlations of 0.1 or greater.
Characteristics of study participants were presented as medians and corresponding interquartile ranges (IQR) for continuous variables and as percentages for categorical variables. Between sex differences were tested using Wilcoxon rank-sum test or chi-squared test.
Due to the highly skewed nature of HCC, this variable was log-transformed to attain a normal distribution. For descriptive purposes, both actual values and log-transformed values are presented in descriptive part of the result section.
Linear regression analyses were conducted to assess the associations between baseline log HCC and subsequent 6, 12 and 18-month change in BW and BF%, in addition to BWV over the same time periods. Moreover, we examined the association between 12-month concurrent changes in HCC and the outcome measures, and the association between initial change in log HCC from baseline to month 12 and subsequent change in outcome measures between month 12 and month 18.
First, crude models, including information on outcome, exposure and baseline measure only were conducted. Secondly, adjusted analyses with added information on initial WL, smoking status, frequency of alcohol consumption, physical activity, education, age, intervention status (arm allocation), centre (country of residence) and sex as potential confounding factors were conducted.
Model assumptions (investigating linearity of effects on outcomes, consistency with a normal distribution, and variance homogeneity) were assessed through visual inspection of histograms and residual plots. Some models with BWV as the outcome measure showed slight deviation from the assumption of normally distributed residual. However, as sensitivity analyses with log-transformed outcomes gave essentially the same results, analyses with non-transformed outcome measures are presented throughout.
Sex and intervention interactions were tested in all analyses by adding product terms to the models, and subgroup analyses were conducted if significant interaction was observed.
All statistical tests were two-tailed with a significance level at 0.05. BWV estimates were calculated in R version 3.4. Analyses were performed using Stata SE 14 (StataCorp LP, College Station, Texas, USA).

Sensitivity Analyses
To get associations independently of self-reported stress, sensitivity analyses additionally adjusted for the PSS were conducted. Moreover, since stress is closely related to sleep (39), and we have previously shown an association between sleep onset variability and weight regain among the NoHoW participants (35), sensitivity analyses were conducted adjusting for sleep duration and sleep onset variability. Analyses adjusted for hair washing frequency and use of hair dye were conducted as well, although the evidence for potential influence on HCC from these factors is inconsistent (21,40). Sensitivity analyses were also performed adjusting for weight of the hair sample, as we did not have 10-30 mg of hair for all individuals and we found that log HCC was positively correlated with weight of the hair samples (r=0.2). All analyses were also conducted excluding participants with less than 10 mg of hair at baseline (n=213) and follow-up (n=74). Finally, we performed sensitivity analyses excluding participants currently treated with any form of medication (n=301)

RESULTS
A total of 172 men and 614 women were included in the study.   and a lower BF% was observed for men than women at all visits (all P<0.001) ( Table 2), while no consistent differences in BWV was observed ( Table 3).
We found no associations between baseline log HCC and subsequent change in BW or BF% during 6, 12 or 18 months. Moreover, while crude models showed baseline log HCC to be associated with higher subsequent BWV during the first 6 and 12 months, using both RMSE and NLMD, no statistically significant associations were observed after adjustment for covariates ( Table 4). Likewise, we found no associations between 12-month concurrent changes in log HCC and BW, BF% or BWV ( Table 5). Finally, while no association between initial 12month changes in log HCC and subsequent change in BW and BF% was found, an initial increase in log HCC was associated with a higher subsequent BWV. However, after adjusting for potential confounding factors this association was only statistically significant using the NLMD method (b: 0.02% per u n i t i n c r e a s e i n l o g H C C [ 9 5 % C I : 0 . 0 0 , 0 . 0 4 ] ; P=0.016) ( Table 6).
None of the analyses showed evidence of interaction between log HCC (baseline level or change) and sex or intervention status (all p-values > 0.05).

Sensitivity Analyses
Additional adjustment for the PSS, sleep duration and/or sleep onset variability gave essentially similar associations (Supplementary Tables 1-3), as did adjustment for weight of the hair sample, use of hair colouring products and hair washing frequency (Supplementary Tables 4-6), indicating no or limited confounding from these factors. Analyses excluding participants with less than 10 mg of hair also gave similar associations (Supplementary Tables 7-9). Likewise, analyses excluding participants treated with any form of medication gave almost identical results (Supplementary Tables 10-12).

DISCUSSION
In a large WLM trial of European men and women, we found no evidence of association between baseline HCC and subsequent change in BW, change in BF% or BWV when controlling for lifestyle and demographic factors. Likewise, no associations between 12-month concurrent development in HCC and outcome measures were observed, but an initial 12-month increase in HCC was associated with a higher subsequent BWV between month 12 and 18. While a large body of evidence has confirmed a crosssectional association between HCC and adiposity (21)(22)(23)(24), to our knowledge, no previous studies have explored the relationship between HCC and WLM. However, similar studies have been conducted based on self-reported information on psychosocial stress. For instance, Brantley and colleagues (2014) found perceived stress to be associated with weight regain in a WLM trial of 1,025 men and women (10), and some intervention studies have suggested that stress management can facilitate WL or weight maintenance (41)(42)(43). However, our results suggest that an increased long-term cortisol secretion was not the primary driver of these associations.
Likewise, we found no previous studies examining the association between HCC and BWV, but results from 4,774 participants from the Look AHEAD study suggested an Results presented as change in outcomes (95% CI) between baseline and month 12 per additional unit of 12-month concurrent change in log hair cortisol. 2 BF%, Body fat percentage; BW, body weight; NLMD, non-linear mean deviation; RMSE, root-mean-square deviation. 3 Model with exposure, outcome and baseline measures of both, only. association between mental health or depressive symptoms and BWV (44). This may explain the association between initial 12 month increase in HCC and subsequent 6 month BWV observed in the present study as van Manen and colleagues (2019) have shown HCC to be directly associated to subjective measures of psychological distress (45). Another explanation for the association between an increased HCC and a subsequent higher BWV observed in our study may be a tendency towards more irregular eating patterns when stress levels increase (13,14). Such an eating pattern could theoretically lead to a higher degree of fluctuation in calorie intake, without necessarily affecting the average intake. However, it is worth mentioning, that although some studies have shown a high degree of BWV to be associated with both future weight gain, morbidity and mortality (16,46), these studies are often based on too few BW measurements to get an accurate estimate of BWV and thus any clinical significance of our results is uncertain (30). The present study has several strengths, including the use of data from a large WLM trial with repeated information on HCC and outcome measures, and the use of prospective analyses which reduces the risk of reverse causality. Moreover, in addition to data collected at the four centre visits, all participants were instructed to weigh themselves on the validated Fitbit Aria scale (27) at least twice per week for the 18 month duration of the study, providing us with detailed measures of BWV. We also had verified information on several lifestyle factors, including objectively measured physical activity and sleep habits, allowing us to adjust for potential influence from these factors.
The study also has some limitations. For the analyses of HCC, we aimed to analyse 10-30 mg of hair. However, especially among the male participants, the collected samples did not consistently contain sufficient hair to meet this criterion. As we found a direct correlation between the weight of the hair samples and HCC, this may have affected the measured values. However, as we found almost identical associations estimates in sensitivity analyses adjusted for weight of the hair sample, and when excluding individuals with less than 10 mg of hair available, it seems unlikely that this had a substantial influence on the observed associations. The detailed BW information, collected twice weekly using the validated Fitbit Aria scale, gave us a detailed measure of BWV during the trial, but we do not know exactly what tissues or circumstances were actually fluctuating (e.g., fat mass, water, glycogen, weight of gut content, clothing and/or time of day), and indeed it is likely that the majority of fluctuating body weight relates to acute fluctuation in total body water (47). In addition, although we adjusted our analyses for several potential confounding factors, we cannot exclude that some unmeasured or residual confounding has remained. As an example, although we did perform sensitivity analyses excluding participants currently treated with any form of medication, which did not change the observed associations, we did not specifically ask participants on their use of corticosteroids. Moreover, we did not adjust for multiple testing, and thus results for the secondary outcomes should be interpreted with caution due to the risk of false positive associations.
We conducted analyses of the association between HCC and outcome measures using three different analytical approaches, as each of them have their own set of limitations. In analyses of baseline HCC and subsequent change in adiposity or BWV, the temporal separation between exposure and outcome measures is a clear strength. However, when using this approach, a comparison of different individuals who vary in HCC is made. As HCC has a high genetic component (48,49) and people are likely to vary according to other factors than their HCC, confounding may be a more pronounced problem when using this strategy. Analysing concurrent 12-month changes in two measures is subject to the same limitations as a cross-sectional design, and thus it is not possible to determine causality. In contrast, analysing the association between preceding changes in HCC during the first 12-month period and outcomes during the subsequent 6-month period is a strength in terms of keeping a prospective structure in the analyses (e.g. temporally separating the change in HCC from the change in outcome measures). Results presented as change in outcomes (95% CI) between month moth 12 and 18 per additional unit change in log hair cortisol between baseline and month 12.
2 BF%, Body fat percentage; BW, body weight; NLMD, non-linear mean deviation; RMSE, root-mean-square deviation. 3 Model with exposure, outcome and baseline measures of both, only. 4 Adjusted for physical activity (steps/day), smoking status, educational level, sex, alcohol, age, height, intervention status, country and initial weight loss.
However, this approach also introduces a limitation, as the potential effect of changes in HCC on adiposity may have occurred in the same 12-month period or shortly thereafter. Finally, our results primarily originate from individuals with overweight or obesity. All subjects had achieved a clinically significant WL prior to enrolment, and ¾ of the participants received a digital intervention based on self-regulation, motivation, and/or emotion regulation tools designed to improve WLM. Thus, generalisation to the general population should be done with caution.
In conclusion, our data suggest that an association between HCC and WLM is either absent or marginal. An initial increase in HCC may predict a slightly higher subsequent BWV, but the clinical importance of this is unknown. Additional longitudinal studies are needed to establish the causal direction and underlying mechanisms of the association between HCC and adiposity.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://easo.org/thenohow-dataset/.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by local institutional ethics committees at the Universities of Leeds  the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Copyright © 2021 Larsen, Turicchi, Christensen, Larsen, Jørgensen, Mikkelsen, Horgan, O'Driscoll, Michalowska, Duarte, Scott, Santos, Encantado, Palmeira, Stubbs and Heitmann. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.